AI Code Quality: Bad Code Is an AI Tax

The model is the brain. The harness is the operating environment that makes the brain useful. A plain-English breakdown of agent loops, tools, permissions, memory, verification, and why harness design matters more than model choice.

Pull requests are up 20%. Incidents per PR are up 23.5%. The supervision gap is real and most managers are missing it.

The gap between Kimi K2.6 and GLM-5.1 on public coding benchmarks is basically noise. The real decision is when to route work to them versus Claude Opus 4.7.

Taste isn't mystical. It's System 2 thinking, run on the same decision so many times it drops into System 1. That's why AI can't get there — and why it's a moat.

A reader asked whether DOE still holds up now that Claude Skills, Managed Agents, and Cowork exist. Short answer: yes, and his question sharpened the framework in ways the original post didn't.

A 2-minute AI readiness quiz for engineering teams using Claude Code, Cursor, Cowork, or ChatGPT. Five questions, four levels, and the one thing to do next at each stage — whether your team ships code, decks, or reports.

A generalist's field notes from 2026. AI changed the game, execution still wins, and the engineers who connect dots nobody else sees are the ones getting ahead.

Your team is calling 4 LLM providers from 6 services with no routing layer. Here's the architecture pattern that fixes it.

A 4-factor framework for evaluating AI automation in service businesses. Real tools, actual cost ranges, and when the math doesn't work.

AI tools made your team faster. Then patterns started drifting. Here's how to keep architectural coherence without killing the productivity gains.

Real automation costs for small businesses: $0-10K. Three tiers and the ROI math that makes the decision obvious.

Individual developer metrics are up. Organizational metrics are flat. Here's why the gap exists and what it means for AI coding tool adoption.

A practical no-code stack for founders: 8 tools in 3 layers, $9-73/month, with a complete lead-to-client workflow. Start automating this week.

Three real automations I use daily: PandaDoc proposals, a GitHub Actions trend scanner, and automated onboarding. Exact tools and code.

Research, topic selection, writing, post-processing — all run by an AI agent with zero human edits. Here's the full pipeline, the four-treatment experiment, and the honest 7/10 score.

Most AI agent tutorials show Python loops and vector databases. Here's what daily agent use actually looks like: terminal-first orchestration, agent files, context architecture, and a knowledge system that feeds all of it.

AI is writing more code. Here's how to review it faster — local agent patterns, CI/CD integration, and the vendor landscape including Greptile, CodeRabbit, and GitHub Copilot.

Built a beta signup pipeline (form to email in 30 seconds) with Modal, Resend, and Google Sheets. 913 lines of Python, 4 hours, $0/month.

Most developers over-invest in prompt engineering and ignore context. Here's the four-layer framework that makes every AI coding session better.

Four major coding models launched in six days. The benchmark gap? 2.6 points. The price gap? 45x. A head-to-head comparison of Opus 4.6, Codex 5.3, GLM-5, and Kimi K2.5.

Vibe coding was the prompt-first era. Agentic engineering is the orchestration-first era. Here's what changed, why it matters, and what you need to learn next.

The 5 automations that save service businesses 10+ hours/week. Concrete tools, setup times, and ROI for agencies and consultants.

Most developers blame the model when AI coding tools produce bad output. The real problem is context. Here's the system I built to fix that.

Not all tasks need the most powerful AI model. Learn how to match model intelligence to task complexity and stop overpaying for sledgehammers when you need scalpels.

AI tooling can speed up enterprise engineering teams, but only with the right guardrails. A practical guide to prompting, context management, governance, and parallelization across Claude, Copilot, and CLI tools.

AI has moved from a shiny new toy to everyday infrastructure. Here's a realistic look at what that means for developers, managers, and business owners.

Classic design patterns still solve modern problems, but only if you use them to fix real friction. Here's how I use them to build maintainable systems without the architecture astronaut fluff.

A practical, evergreen guide for beginners and no-code users who need reliable CRM integrations across finance, real estate, and professional services.

Structured outputs turn LLM text into dependable, validated data. Learn schemas, validation loops, provider-native features, and practical patterns for extraction, routing, and ETL.

Explore how LLM-powered CLI agents simplify self-hosting on VPS and homelabs. Learn deployment patterns, Docker Compose examples, and guardrails for safe automation.

Cloud-native system design that scales and fails well. Learn durable patterns—consistency, resilience, observability—with lessons from Amazon, Google, Netflix, LinkedIn, and Stripe.

What AI coding assistants actually do well, where they fail, and how to use them without breaking things or forgetting how to code.

A practical guide to BGP anomalies: taxonomy, detection signals, and mitigation patterns. Deploy RPKI, build guardrails, and design networks that limit blast radius.

Sharing lessons learned as a Senior Software Developer.

A breakdown of production-ready AWS Lambda and SQS configurations, covering visibility timeouts, batch sizes, failure handling, and idempotency strategies.

Use DBeaver to run EXPLAIN/ANALYZE, find index gaps, apply VACUUM wisely, and enlist AI to build a safe, testable Postgres tuning plan.

A pragmatic guide to JPA/Hibernate for tech leaders and engineers. Learn hidden tradeoffs, when to use native SQL, and patterns to ship fast, safe data access.

Microservices are not the default. Explore universal-interface architecture and the modular monolith, with concrete patterns, migration steps, and metrics.

Why IaC matters, how to run Terraform at team scale, and a step-by-step EC2 example to get started.

Learn the Model Context Protocol (MCP): what it is, why it matters, and how to ship secure, observable AI tooling with a step-by-step example. Get started now.

Learn n8n: what it is, why it matters, and how to ship secure, observable automations with a free RSS-to-Slack example. Start building value fast.

A pragmatic guide to prompt engineering for tech leaders and engineers. Learn durable patterns, ROI-focused choices, and a full before/after example.

AI now writes a large share of new code. Stand out with Architecture as Code—pros, cons, tools, and a simple adoption plan for leaders, engineers, and recruiters.

How AI is actually changing software development—from what I've seen in the wild and what's working versus what's hype.

How low-code and no-code tools actually work in enterprise environments, based on what I've seen.